Using AI Research Tools to Challenge Support for Police Violence in Brazil

• A Brazilian civil society group, Instituto Fogo Cruzado, partnered with Future Narratives Lab to use an AI-based interviewer tool to research public support for police violence in favelas.
• The AI tool successfully conducted 28 in-depth interviews with a hard-to-reach, pro-police-raid demographic, uncovering key insights into their underlying beliefs.
• The study found that pragmatic messaging regarding police incompetence and lack of training was more effective than traditional empathy-based narratives.
• Findings revealed that Brazilian Evangelical participants often hold nuanced views centered on redemption and distrust of elites, challenging common assumptions about their political alignment.
• The research methodology has been shared with 20 other Brazilian organizations to prepare for the October 2026 general election.

Research Methodology and AI Application

• The AI Interviewer tool enabled researchers to compress weeks of fieldwork into days at a significantly lower cost.
• Participants were selected based on demographic profiles—specifically lower-income, non-higher education, and Christian—that typically support aggressive policing.
• The tool engaged in structured but adaptive 30-minute conversations to probe inconsistencies and reasoning behind participant views on crime and police tactics.
• To address ethical concerns, the researchers focused on privacy-protecting, open-source solutions rather than standard "frontier" AI models.

Findings and Strategic Implications

• The most effective counter-narrative identified was: "The police aren’t sufficiently trained or prepared, so carrying out raids in crowded favelas only leads to more shootouts and more innocent victims."
• The research showed that targeting police "incompetence" resonated more than focusing on structural injustice or victim empathy.
• Future efforts aim to apply these insights to shift public opinion on security policy in the lead-up to the 2026 elections.